To make use of GPT-3 models to answer questions based on a given context, we could fine-tune a model with question-answer pairs, then query the model. Is this the recommended approach? What if we have large number of documents? How can we know from where the model has come up with the answer, say for example, from which document did you get this info?